nodejs self-hosted nvidia multi-model inference-server homelab multi-gpu rocm load-balancing model-serving llm llama-cpp llm-inference local-ai gguf llama-server llm-orchestration openai-compatible gpu-pinning llamafleet